A research paper co-authored by Yosuke Kishinami, Ryo Fujii, and Mutsumi Morishita—Research Engineers in the AI Strategy Promotion Group at Future Corporation (Shinagawa-ku, Tokyo; President and CEO: Tomohiko Taniguchi; hereinafter "Future")—has been accepted to the Main Conference of "ACL2026" (Association for Computational Linguistics 2026) (Note 1), a premier international conference in the field of natural language processing. ACL2026 (The 64th Annual Meeting of the Association for Computational Linguistics) is a conference hosted by the Association for Computational Linguistics (ACL), the world's largest academic society in the field of natural language processing. The accepted paper will be presented at ACL2026, to be held in San Diego, USA, from July 2 to July 7, 2026. This research is a collaborative effort with Nara Institute of Science and Technology, a national university corporation. ■ Accepted Paper Revisiting Non-Verbatim Memorization in Large Language Models: The Role of Entity Surface Forms Yuto Nishida, Naoki Shikoda, Yosuke Kishinami*, Ryo Fujii*, Makoto Morishita*, Hidetaka Kamigaito, Taro Watanabe *Affiliated with Future Corporation https://aclanthology.org/2026.acl-long.2178.pdf 【Research Overview】 Large language models (LLMs) store various kinds of knowledge. This study investigates whether knowledge memorized by LLMs can be equally retrieved regardless of the name used—such as official names, aliases, or abbreviations. To evaluate this, we constructed a new evaluation dataset called RedirectQA, using alternative mentions of people and organizations from Wikipedia, and tested multiple LLMs. The results show that LLM responses often vary depending on the name used to refer to the same entity. This reveals that robust evaluation of LLMs must consider their ability to handle diverse naming variations. At Future, we are actively cultivating and hiring research engineers and AI engineers, primarily through our specialized AI organization, the AI Str